handleSubmit's ~200-line if/return chain becomes a declarative
{test, run}[] list plus one dispatch loop. Same matching semantics
(exact vs. loose startsWith), same order (/permissions before /perm,
/model before /mode), same handler bodies — just uniform structure and
a single place the command set is registered. Handlers still close over
component state; extracting them into an independently testable module
is a larger follow-up.
Also drops a stale comment claiming Ollama can't report context length
for ":cloud" models — /api/show now returns it for every cloud model
tested (glm-5.2, qwen3.5, kimi-k2, minimax, gemma4).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
⏱ (U+23F1) and ⏵ (U+23F5) are text-default emoji: string-width counts
them as 1 column but Windows Terminal, iTerm and others render them as
2. That makes a status-bar line drift wider than Ink's layout math
expects, so the live region below it under-erases on re-render and
leaves stranded copies of the bar in scrollback. Swap in ⏳ (a full RGI
emoji, unambiguously 2) and ▸ (a plain geometric shape, unambiguously 1).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
tsx 4.23.13 pulls its own esbuild 0.28.1; allowScripts only listed
0.27.2 (tsup/vite), so npm warned about the blocked postinstall. Also
records the marked<16 / typescript-7 / wrap-ansi-10 constraints in the
memo so a future upgrade pass doesn't have to rediscover them.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
tsc --noEmit passes with zero errors and runs noticeably faster. The
project only uses the tsc CLI for typechecking (no programmatic
typescript API imports); tsup/esbuild drives the build and is
unaffected. The native port pulls per-platform @typescript/typescript-*
binaries as optional deps, like esbuild already does.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Dev-only. Config (globals/environment/include) and the vi.* API surface
used by the suite are unchanged; all 45 files / 366 tests pass. Node 24
satisfies vitest 5's engine range.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The previous commit's cache key used a raw NUL as the width/text separator,
which made git track src/ui/render.ts as a binary file. Use a plain-text
"<width> <text>" key instead (width is always a bare integer, so the space
delimiter is unambiguous) and add renderMarkdown width-wrapping tests.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Lines wider than the terminal (wide markdown tables, long code lines) were
soft-wrapped by the terminal but counted as one row by Ink's <Static> redraw,
desyncing its cursor math and stranding duplicate copies of the status bar and
input box in scrollback on every re-render.
renderMarkdown now wraps its output to the caller's width via wrap-ansi, and
every <Static> item (plus the live streaming line) is held within a
width-bounded Box so nothing the terminal would soft-wrap reaches Ink.
Also sets cli-table3 header color to yellow (red-on-blue themes made table
headers unreadable).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
locode-authored (dogfooded) changes:
- TaskStore.toJSON/fromJSON: task_create/list state now survives session
resume instead of resetting to empty (session.ts, task.ts, sessionStore.ts).
- StatusBar/ChatInput/ThinkingIndicator wrapped in memo() to skip re-renders
during the ~30fps streaming-text update loop.
- HistoryItemView's todos list keys off t.content instead of array index
(TodoItem has no id field; content is the best stable key available).
- sessionStore.ts read paths now log (console.warn) on corrupt/unreadable
files instead of silently swallowing the error.
Fixes on top, found in review:
- safeRunHooksForEvent() wraps the 4 remaining unprotected hook call sites
(PreToolUse, PermissionRequest, PostToolUse, SubagentStart) — these fire on
every tool call/sub-agent, far more often than the 3 sites already wrapped
(Stop/SessionStart/UserPromptSubmit), so a broken hook script was still able
to abort an in-flight turn.
- deleteSession() no longer reports success when unlinkSync fails for a
reason other than ENOENT (e.g. EPERM/EBUSY from a Windows file lock) — it
used to drop the index entry and return true anyway, orphaning the file on
disk with no way to reference it again. `sessions rm` in cli.ts now
distinguishes "not found" from "found but couldn't delete" in its message.
- ChatInput's memo() was a no-op: onSubmit={handleSubmit} passed a fresh
closure every render since handleSubmit isn't wrapped in useCallback.
Rather than force a ~300-line function (which reads isThinking/streamingText
— both changing every streaming frame) into a dependency array, added a
handleSubmitRef + stable wrapper, the same ref-indirection this file
already uses for phaseRef.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LxiAaGhSD4DRVYYQZ5GJjm
- isSmallLocalModel(baseURL, model) replaces isLocalBackendURL() for system-prompt
branching and context-window fallback defaults: Ollama's cloud-routed models
(glm-5.2:cloud, qwen3.5:397b-cloud, etc.) share a localhost endpoint with
genuinely local models, so the base URL alone can't tell them apart. Recomputed
on /model and /backend switches too, not just at session creation.
- Fixed a latent isLocalBackendURL bug found while testing it: URL.hostname keeps
the brackets on a literal IPv6 host ("[::1]"), so the old "::1" comparison never
matched.
- capabilityCache entries now carry a cachedAt timestamp with a 30-day TTL
(LOCODE_CAPABILITY_CACHE_TTL_DAYS), so a stale "fallback" verdict from a
transient probe failure doesn't permanently disable native tool calls.
- ChatInput's cursor row offset (+2 -> +1): the extra row was empirical padding
for a bottomSectionRef wrapper Box and virtual-scroll viewport that no longer
exist since the Static-based rendering change.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LxiAaGhSD4DRVYYQZ5GJjm
Major changes:
- Replace alternate screen buffer + app-side mouse tracking/virtual scroll
with normal screen buffer + Ink <Static> for permanent scrollback.
Terminal's native scroll/selection/copy just works — no mouseInput.ts needed.
- Fix burn rate (🔥) to use model response time (modelTimeMs) instead of
session elapsed time, so it reflects actual generation throughput.
- Upgrade dependencies: openai 6→7, commander 13→15, execa 9→10, vitest 3→4,
node types 22→26, tsup target node20→node22.
- Update upgrade memo with current architecture, all completed upgrades,
and working tree status.
- App.tsx: add the missing effectiveScrollTopRef the prior session's mouse-drag
fix referenced but never declared — tsc failed (esbuild build doesn't
type-check, so it looked fine), and any click/drag/release threw
ReferenceError at runtime.
- partialJson.ts/nativeAdapter.ts: escape raw control characters (literal
newline/CR/tab) found inside JSON string literals before parsing tool-call
arguments, and add repair to the non-streaming completion path (previously
only the streaming path had it) — local models echoing multi-line CRLF file
content unescaped were failing JSON.parse outright.
- editFile.ts/multiEdit.ts: read_file always shows the model LF-normalized
content regardless of the file's real line endings, but edit_file/multi_edit
matched old_string against the raw (CRLF-preserving) file — a systematic
"old_string not found" on every CRLF file in this project. Now detect the
file's EOL, match/edit in LF space, and restore the original EOL on write.
- StatusBar.tsx: 🔥 burn rate now uses outputTokens/min instead of
(inputTokens+outputTokens)/min — inputTokens sums the full resent context on
every API call (no prompt caching), so it ballooned into misleading hundreds
of K/min during chatty fallback-mode tool round trips.
339/339 tests passing (+13), tsc clean, build 295.77 KB.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Kn75SyL8t8SeyTwmPvADAS
- Separate system prompt for local vs cloud models via isLocal flag
(isLocalBackendURL, buildSystemPrompt(..., isLocal), propagate through
createSession/createSessionFromRecord/compactSession/spawnSubAgent)
- Increase MAX_EMPTY_RESPONSE_RETRIES from 1 to 3 for cloud model resilience
- Upgrade mouse input: full SGR-1006 parsing with col/row/pressed/modifiers,
logicalButton() helper, and copyToClipboard() via OSC 52
- Add temporary mouse debug logging in App.tsx
- Increase DEFAULT_MAX_ITERATIONS from 100 to 300
- Update README mouse/scrollback docs, tweak diff-remove color
While a turn was in flight (isThinking or streamingText), the ChatInput was
unmounted and replaced with a static "Waiting for response…" box — so the user
couldn't type ahead during output, the input field just vanished.
Now ChatInput stays mounted the whole time: the user can type and edit while
output streams in. A one-line status above it ("Waiting… esc to interrupt — type
ahead, Enter sends when done.") replaces the old full-width box. Submitting a
second turn mid-stream is still prevented — handleSubmit now guards on
isThinking/streamingText and returns early, so Enter is a no-op until the
in-flight turn finishes (the typed text stays in the box). The original race this
swap was protecting against (overlapping turns on the same session) is preserved
by that guard.
Verified: typecheck clean, build 274.76 KB, 301 tests pass.
Replaces the flat todo_write approach for non-trivial multi-step work with a
structured, incrementally-updated task store — dependencies, ownership, status,
metadata — ported from origin/v0.6.0 and adapted to the local Session/ctx.
- src/tools/task.ts: TaskStore (in-memory, per-session) + four read-only tools:
task_create (returns id), task_list (summaries), task_get (full details),
task_update (status pending|in_progress|completed|deleted, rename, claim via
owner, addBlocks/addBlockedBy dependency edges, merge-patch metadata — null
deletes a key). Self-refs/unknown ids ignored; direct 2-cycles skipped;
delete prunes dangling refs. Schemas declared before use (fixes a TDZ in the
v0.6.0 version). todo_write kept alongside for simpler cases.
- types.ts: ctx.taskStore. session.ts: TaskStore on every Session. loop.ts:
expose session.taskStore in the tool ctx + on sub-agent sessions.
- tools/index.ts: register the four task tools.
- task.test.ts: 9 cases (create/list/get/update status, dependency edges both
ways, self-ref/unknown/2-cycle guards, delete+prune, metadata merge-patch,
absent-store error).
- README: document the structured tasks.
Verified: typecheck clean, build 275.10 KB, 301 tests pass (+9).
The markdown export used to include only plain user/assistant text — every tool
call and tool result was dropped, so a shared transcript lost half the work.
- exportSession.ts: the markdown export now includes tool calls (fenced JSON),
tool results (fenced), fallback tool_result blocks, and multimodal user
content (text + [image attached] placeholders). Adds sessionToJson() — the
full record (messages verbatim + meta + exportedAt) for cross-machine
replay/sharing. exportSession() takes a format; defaultExportFilename() picks
.md/.json.
- App.tsx: /export json [file] selects the JSON dump; /export [file] stays the
markdown transcript. The export prompt remembers the chosen format.
- exportSession.test.ts: 8 cases (markdown includes tool calls/results, tool-
only assistant turn, JSON shape, filename extensions, file writes, auto-name).
- /help + README: document /export json.
Verified: typecheck clean, build 268.34 KB, 292 tests pass (+8).
Ported from origin/v0.6.0, adapted to resolveWithinCwd (no setLastEdit).
- src/tools/notebookEdit.ts: replace/insert/delete a cell by cell_id or
cell_index. Converts the model's single-string new_source to/from nbformat's
source line-array (trailing-newline convention), so the model never hand-writes
the array quirk. Switching a code cell to markdown drops execution_count/outputs;
switching to code adds them. Atomic temp+rename write, JSON validated.
- tools/index.ts: register notebook_edit.
- notebookEdit.test.ts: 9 cases (replace by id/index, insert at position/append,
delete, missing-id failure, insert-without-cell_type, replace-without-
new_source, diff preview, code→markdown field drop, path containment).
Prefer this over edit_file/write_file for .ipynb so the JSON structure stays
valid (and avoids regenerating a whole notebook JSON in the model's output —
a context-window win for local models).
Verified: typecheck clean, build 266.91 KB, 284 tests pass (+9).
The pre-confirmation preview for mutating tools (edit_file/multi_edit/write_file/
bash/git_commit) was rendered as plain monochrome text, throwing away the most
useful signal in the thing the user evaluates before approving.
- src/utils/diff.ts: looksLikeDiff() — distinguishes a unified diff from a plain
preview (bash command, 'Create new file ...', JSON args).
- src/ui/ink/DiffView.tsx: renders a unified diff with additions green, removals
red, hunk/file headers dimmed. Optional side-by-side mode (pairHunk + clip,
both exported for testing). Ported from origin/v0.6.0.
- src/ui/theme.ts: DIFF_ADD_HEX / DIFF_REMOVE_HEX.
- PermissionPrompt.tsx: use DiffView when the preview is a diff, keep the
line-by-line render for non-diff previews.
Tests: diff.test.ts (4), DiffView.test.ts (8) for pairHunk/clip/looksLikeDiff.
Verified: typecheck clean, build 261.50 KB, 275 tests pass (+12).
Mouse-wheel tracking (xterm ?1000h) was always on so the wheel could scroll
the transcript, but it also captures mouse events, which prevents the
terminal's native text selection/drag-to-copy — a frequent ask.
Now mouse tracking defaults to OFF, so selecting/dragging output to copy
works out of the box. PageUp/PageDown still scroll (they always did). Toggle
the wheel back on with '/mouse on' (re-captures mouse events, so selection
won't work while on); '/mouse off' restores selection. '/mouse' shows the
current state. Documented in the /help list and README.
Verified: typecheck clean, build 257.55 KB, 263 tests pass.
The built-in LANGUAGE_SPECS were hardcoded — users couldn't add Java/Ruby/Lua
servers or override a built-in's command/args, and the file's own comment
flagged this as a future extension. Also, c and cpp were separate specs both
spawning clangd, so a mixed C/C++ project ran two indexing the same headers.
- lspManager.ts: LANGUAGE_SPECS is now mutable; configureLanguageSpecs(overrides)
merges user entries (keyed by languageId) into the built-ins. A built-in id
override replaces command/args and, if extensions is given, rewrites routing.
A new id adds a mapping but REQUIRES extensions (ignored otherwise — can't
route files to it). C and C++ collapse into one 'c' clangd spec (all
.c/.h/.cpp/.cc/.cxx/.hpp/.hh/.hxx route to a single clangd).
- config.ts: resolveLspServers() reads stored.lspServers.
- store.ts: StoredConfig.lspServers field.
- cli.ts: 'locode config set lspServers <json>' (JSON object value, validated).
- ui/ink/index.tsx: configureLanguageSpecs(resolveLspServers()) at startup.
- lspManager.test.ts: 6 tests for the merge (add, override, rewrite exts,
ignore-without-extensions, c/cpp collapse) via _specsForTests/_resetSpecsForTests
— no servers spawned.
Verified: typecheck clean, build 256.60 KB, 256 tests pass (+6).
getDiagnostics used to sync the document then wait one macrotask
(setTimeout 0) before reading the cached snapshot. tsserver/pyright on a
large file publish asynchronously and often hadn't fired yet, so the call
returned a stale (or empty) snapshot right after an edit — exactly when the
model asks for diagnostics to verify its change.
Now: clear the stale URI snapshot, sync, then race the next
publishDiagnostics for that URI against a 1500ms timeout via a per-URI
waiter map woken by the publish handler. A slow server gets a real chance
to compute fresh diagnostics; on timeout we fall through to whatever's
cached (possibly empty). _resetForTests clears the waiter map too.
Verified: typecheck clean, 250 tests pass.
Ported from origin/v0.6.0 and adapted to the local codebase (resolveWithinCwd
path guard, no setLastEdit). Reuses edit_file's applyEdit/countOccurrences.
- src/tools/multiEdit.ts: applies an ordered batch of {old_string, new_string,
replace_all?} edits to one file. Each edit validates against the running
result (an earlier edit can shift the text a later edit matches), so a
mismatch names its edit index. One confirmation + one temp+rename atomic
write instead of N edit_file round-trips — the biggest local-model win,
since each edit_file is a full generation + permission prompt.
- editFile.ts: export applyEdit + countOccurrences for reuse.
- tools/index.ts: register multi_edit alongside edit_file.
- multiEdit.test.ts: 8 cases (ordered batch, chained edits, error on first
mismatch w/ index, ambiguity guard, replace_all, path containment, diff
preview, preview warning).
Verified: typecheck clean, build 254.81 KB, 250 tests pass (+8).
Three upgrades targeting the most common local-model failure modes where a
tool call is intended but never executes:
#2 Partial/truncated-JSON recovery (native streaming):
- src/toolcalling/partialJson.ts: cheap structural repair for tool-call
arguments that fail JSON.parse — close unterminated strings, balance
open braces/brackets (max_tokens truncation), strip trailing commas and
stray trailing tokens. Never invents keys/values; a repaired call still
goes through schema validation.
- agent/loop.ts: on a parse failure in accumulated native tool calls, try
repair before falling back to a full non-streaming regeneration (which
is expensive on a local backend and fails identically when the cause was
max_tokens). Repaired args replace the broken ones; only unrepairable
calls trigger the retry.
- partialJson.test.ts: 10 cases (truncation, trailing comma nesting,
braces inside strings, escaped quotes, stray trailing content).
#3 Fallback tool-call parser robustness:
- src/toolcalling/fallbackParser.ts: now accepts ```tool_call blocks
(multiline-anchored so an inner ```json fence isn't read as the
terminator), ```json blocks, AND bare unfenced tool-call objects in
prose. Inner ```json fences are stripped. Every candidate runs through
partial-JSON repair, so a truncated fence (no closing ```) still
recovers. Only accepts bare braces that contain "name"+"arguments" keys.
- fallbackParser.test.ts: 11 cases.
#6 Empty-response retry:
- agent/loop.ts: a bare empty `stop` (no text, no tool calls — common from
small/quantized local models) now retries once with a nudge instead of
returning "" or hard-erroring. After the retry budget is exhausted, a
genuine empty response throws "Empty response from model."
- loop.test.ts: retry-then-succeed and retry-exhausted-throws cases.
Verified: typecheck clean, build 251.48 KB, 242 tests pass (+22).
- Switch to the terminal's alternate screen buffer once a session starts,
clearing and homing the cursor so content starts flush at (0,0)
- Replace <Static> scrollback with a fixed-height viewport (overflowY hidden)
that top-aligns short conversations and auto-scrolls to the latest message
once content exceeds the screen, keeping the input pinned to the bottom
- Rewrite ChatInput's cursor handling: track cursor offset directly instead of
relying on ink-text-input, and report the real terminal cursor position via
useCursor + a Yoga-tree walk (absolutePosition.ts) so CJK IME composition
windows anchor at the actual caret instead of the terminal's fallback corner
- Print a "resume this conversation" hint after leaving the alt screen on exit
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Every exit path (Ctrl+C, /exit, SIGTERM/SIGHUP) now prints "Resume this
conversation later with: locode --resume <id>" once Ink has released the
terminal, so the session id isn't lost the moment the process ends.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- Add todo_write tool with live checklist rendering in the UI
- Add permission modes (default/plan/auto-edit/auto-accept) cycled via Shift+Tab or /perm
- Load CLAUDE.md/AGENTS.md as project instructions into the system prompt
- Forward sub-agent tool calls/results to parent UI panel
- Auto-compact context mid-turn and keep model moving with a synthetic user turn
- Prune older image_url parts to cap vision token cost
- Idle-abort guard for stalled streaming backends
- Improve background bash jobs, kill whole process trees on timeout/exit
- Expand hooks: command/http hooks, blocking, JSON outputSchema
- MCP: JSON schema → Zod conversion, parallel connection, duplicate server detection
- Add tests for loop, confirmFn, projectInstructions, readFile, bash, todoWrite, contextWindowCache
A day of dogfooding locode on itself surfaced several real bugs, most centered on
the backend request lifecycle silently hanging with no error surfaced to the user:
- bash/hooks: execa's shell:true defaulted to cmd.exe on Windows, which lacks
Unix tools (cat, sed, grep, ...) and fails a whole pipe with an unhelpful exit
255 if any stage isn't found. Added src/utils/shell.ts to resolve real Git Bash
when available.
- agent/loop: the backend client's own `timeout` only bounds time-to-first-byte —
once a streaming response starts, a backend that goes silent mid-stream (or one
that sends periodic content-less "heartbeat" chunks to keep the connection alive
through a proxy) hung the request forever with zero output. Added an idle-abort
guard that pokes only on chunks carrying real progress (usage/text/tool-calls/
finish_reason), covering the main stream, the non-streaming malformed-JSON
retry, and compactSession.
- agent/loop: a single tool-call-heavy turn (e.g. reading dozens of files) had no
auto-compaction safety net at all — shouldAutoCompact was only ever checked
between turns, never during one, so a long turn could blow past the context
window with no mid-turn correction. Now checked every iteration; re-anchors
mutationCommitLength afterward since compaction replaces session.messages
wholesale.
- agent/loop: hitting the iteration cap surfaced as an opaque "Request failed"
even when every prior tool call (including file edits) had actually succeeded.
Added MaxIterationsError, rendered as a calm "paused, send another message to
continue" notice instead, and raised the default cap 25 -> 50.
- agent/loop: removed a dead-code branch (fallback-malformed retry handling
placed inside a native-mode-only code path, so session.mode === "fallback"
could never be true there) that TypeScript correctly flagged as a type error.
- ui/App: text_done fires with an empty fullText for pure tool-call turns (no
preceding prose); the assistant-message push had no guard, rendering a blank
line every time. Now skipped when fullText is empty.
- ui/App+index: alternate-screen-buffer mode and StatusBar-above-ChatInput
layout, fixing the reported "prompt starts below the dashboard, jumps up on
space" glitch. Added input history (up/down recall).
- utils/writeFileAtomic: random temp-file suffix instead of a fixed `<file>.tmp`,
so concurrent writes to the same directory can't collide.
Also fixes two hardcoded version strings (cli.ts --version, mcp/client.ts's
self-reported MCP identity) that were left stale at 0.2.0 through the 0.3.0
bump, and a stale maxIterations default noted in the README.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Bump to 0.3.0.
- bash_kill tool lets the model terminate a still-running backgrounded
job (Ctrl+B); all running jobs are killed on process exit so they
don't outlive locode as orphans.
- Finished background jobs are now evicted from the registry after a
10-minute TTL instead of accumulating for the life of the session.
- The chat completion request timeout is now configurable
(requestTimeoutMs / LOCODE_REQUEST_TIMEOUT_MS, default 180000) so
backends that queue requests behind a concurrency limit (e.g.
Ollama's OLLAMA_NUM_PARALLEL) under multi-session load aren't forced
into a hard failure at a fixed 3 minutes.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Builds out locode's Claude Code plugin parity: MCP servers, slash commands,
sub-agents, hooks (12 lifecycle events), and skills, all loadable from a
local path or git URL. Also adds image support (read_file, /import), @-mention
file autocomplete, a /dashboard stats view, /export with an editable filename
prompt, an expanded git tool (reset/stash/merge/rebase/delete_branch), and a
redesigned status bar.
On top of that, a review pass found and fixed 10 correctness/stability bugs:
argument-injection in git reset/merge/rebase (a ref like "--hard" was parsed
as a flag), tool-call image results interleaving with native tool_call_id
messages and breaking OpenAI-compatible message ordering, backgrounded bash
jobs still being silently killed by their original timeout with the kill
masked as a clean exit, a SubagentStart hook's block being ignored, the
sub-agent timeout clock starting before the hook it should exclude, a
template-expansion bug that could re-substitute $1..$9 placeholders,
unbounded background-job output buffers, a missing directory-target fallback
in /export, and image size caps checked after reading the whole file instead
of before.
Session saves and exports now go through a shared atomic
write-then-rename helper so a crash can't leave a truncated file.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-07-03-134954.txt was a stray captured session log containing an
email address, tracked since the initial commit. Ignore future files
matching the same timestamped naming pattern.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The 8-iteration-per-turn cap was hardcoded from before this session's feature work and
too tight for genuine multi-file tasks, causing "Max tool-call iterations reached" on
legitimately multi-step work. Raised the default to 25 and made it configurable via
`locode config set maxIterations <n>` / $LOCODE_MAX_ITERATIONS, mirroring the existing
contextWindow config pattern.
Also bumps the version to 0.2.0 (package.json, --version, and the MCP client's
self-reported identity) to reflect the substantial feature additions since 0.1.0.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Substantially expands locode's tool loop toward feature parity with Claude Code:
- web_search / web_fetch: DuckDuckGo-backed search and URL fetching (HTML stripped to text)
- git_status / git_commit: structured git tools with real diff/status/commit-list previews
before any mutating operation is confirmed
- agent tool: sub-agents run an isolated, headless tool loop sharing the parent's permissions;
bounded by an explicit depth guard (no nested sub-agents) and a 120s hard timeout, independent
of the toolset already excluding `agent` for sub-agent sessions
- MCP client support (stdio + streamable HTTP transports) via the official SDK: locode connects
to servers configured in `.mcp.json` or via `locode mcp add`, and exposes their tools alongside
the built-in ones, namespaced as `mcp__<server>__<tool>`
- Session persistence: conversations auto-save after every turn; `--continue`/`--resume`,
`locode sessions list/rm`, and an interactive resume picker
- Context compaction: tracks real token usage via `stream_options.include_usage` (falling back to
a char-based estimate), auto-detects the model's context window (Ollama/LM Studio native APIs,
cached, else a configurable default), and auto-summarizes the conversation at 85% usage or on
demand via /compact
- Status bar and welcome banner: added a context-usage indicator and fixed several dim-by-default
colors that were hard to read
Session/loop architecture changed to support all of this: Session now carries its own toolset
(tools + registry + OpenAI schemas) built from local + MCP tools, so runTurn no longer depends on
a fixed global tool list — the same mechanism that lets sub-agents inherit MCP tools while
excluding `agent` itself.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
kimandClaude Sonnet 5